U.S. patent application number 12/631131 was filed with the patent office on 2011-06-09 for system and method for improved automatic speech recognition performance.
This patent application is currently assigned to AT&T Intellectual Property I, L.P.. Invention is credited to Mazin GILBERT, Andrej LJOLJE.
Application Number | 20110137648 12/631131 |
Document ID | / |
Family ID | 44082878 |
Filed Date | 2011-06-09 |
United States Patent
Application |
20110137648 |
Kind Code |
A1 |
LJOLJE; Andrej ; et
al. |
June 9, 2011 |
SYSTEM AND METHOD FOR IMPROVED AUTOMATIC SPEECH RECOGNITION
PERFORMANCE
Abstract
Disclosed herein are systems, methods, and computer-readable
storage media for improving automatic speech recognition
performance. A system practicing the method identifies idle speech
recognition resources and establishes a supplemental speech
recognizer on the idle resources based on overall speech
recognition demand. The supplemental speech recognizer can differ
from a main speech recognizer, and, along with the main speech
recognizer, can be associated with a particular speaker. The system
performs speech recognition on speech received from the particular
speaker in parallel with the main speech recognizer and the
supplemental speech recognizer and combines results from the main
and supplemental speech recognizer. The system recognizes the
received speech based on the combined results. The system can use
beam adjustment in place of or in combination with a supplemental
speech recognizer. A scheduling algorithm can tailor a particular
combination of speech recognition resources and release the
supplemental speech recognizer based on increased demand.
Inventors: |
LJOLJE; Andrej; (Morris
Plains, NJ) ; GILBERT; Mazin; (Warren, NJ) |
Assignee: |
AT&T Intellectual Property I,
L.P.
Reno
NV
|
Family ID: |
44082878 |
Appl. No.: |
12/631131 |
Filed: |
December 4, 2009 |
Current U.S.
Class: |
704/231 ;
704/E15.001 |
Current CPC
Class: |
G10L 15/285 20130101;
G10L 15/32 20130101; G10L 15/00 20130101 |
Class at
Publication: |
704/231 ;
704/E15.001 |
International
Class: |
G10L 15/00 20060101
G10L015/00 |
Claims
1. A computer-implemented method of improving automatic speech
recognition performance, the method causing a computing device to
perform steps comprising: identifying idle speech recognition
resources; establishing a supplemental speech recognizer on the
idle resources based on overall speech recognition demand;
performing speech recognition on speech received from the
particular speaker in parallel with the main speech recognizer and
the supplemental speech recognizer; combining results from the main
and supplemental speech recognizer; and recognizing the received
speech based on the combined results.
2. The method of claim 1, wherein a scheduling algorithm tailors a
particular combination of speech recognition resources.
3. The method of claim 2, wherein the scheduling algorithm releases
the supplemental speech recognizer based on increased demand for
speech recognition.
4. The method of claim 1, the method further causing the computing
device to establish extra speech recognizers tailored to speech
which requires additional accuracy as extra speech recognition
resources become idle.
5. The method of claim 1, the method further causing the computing
device to return the supplemental speech recognizer to an idle
state after recognizing all the received speech.
6. The method of claim 1, wherein the idle speech recognition
resources include networked computing devices.
7. The method of claim 1, the method further causing the processor
to establish a plurality of supplemental speech recognizers on the
idle resources.
8. The method of claim 7, wherein the plurality of supplemental
speech recognizers differ from each other based at least on one or
more of spectral analysis in a front end, pronouncing dictionaries,
and training algorithms.
9. A system for improving automatic speech recognition performance,
the system comprising: a processor; a module controlling the
processor to identify idle speech recognition resources; a module
controlling the processor to establish a supplemental speech
recognizer on the idle resources based on overall speech
recognition demand; a module controlling the processor to perform
speech recognition on speech received from the particular speaker
in parallel with the main speech recognizer and the supplemental
speech recognizer; a module controlling the processor to combine
results from the main and supplemental speech recognizer; and a
module controlling the processor to recognize the received speech
based on the combined results.
10. The system of claim 9, wherein a scheduling algorithm tailors a
particular combination of speech recognition resources.
11. The system of claim 10, wherein the scheduling algorithm
releases the supplemental speech recognizer based on increased
demand for speech recognition.
12. The system of claim 9, the system further comprising a module
controlling the processor to establish extra speech recognizers
tailored to speech which requires additional accuracy as extra
speech recognition resources become idle.
13. The system of claim 9, the system further comprising a module
controlling the processor to return the supplemental speech
recognizer to an idle state after recognizing all the received
speech.
14. The system of claim 9, wherein the idle speech recognition
resources include networked computing devices.
15. The system of claim 9, the system further comprising a module
controlling the processor to establish a plurality of supplemental
speech recognizers on the idle resources.
16. The system of claim 15, wherein the plurality of supplemental
speech recognizers differ from each other based at least on one or
more of spectral analysis in a front end, pronouncing dictionaries,
and training algorithms.
17. A computer-readable storage medium storing instructions which,
when executed by a computing device, cause the computing device to
perform automatic speech recognition, the instructions comprising:
identifying idle speech recognition resources; establishing a
supplemental speech recognizer on the idle resources based on
overall speech recognition demand; performing speech recognition on
speech received from the particular speaker in parallel with the
main speech recognizer and the supplemental speech recognizer;
combining results from the main and supplemental speech recognizer;
and recognizing the received speech based on the combined
results.
18. The computer-readable storage medium of claim 17, wherein a
scheduling algorithm tailors a particular combination of speech
recognition resources.
19. The computer-readable storage medium of claim 18, wherein the
scheduling algorithm releases the supplemental speech recognizer
based on increased demand for speech recognition.
20. The computer-readable storage medium of claim 17, the
instructions further comprising establishing extra speech
recognizers tailored to speech which requires additional accuracy
as extra speech recognition resources become idle.
Description
BACKGROUND
[0001] 1. Technical Field
[0002] The present disclosure relates to automatic speech
recognition (ASR) and more specifically to improving performance of
ASR using idle recognition resources.
[0003] 2. Introduction
[0004] Currently, speech recognition applications are configured as
a compromise between many opposing interests, such as high enough
accuracy, low enough computational load, low enough latency, etc.
One significant element of these opposing interests is the hardware
and computing resources necessary to perform the speech
recognition. Speech recognition systems provision hardware to deal
with peak load demands which may occur at some regular interval or
in extreme situations. However, this approach leaves a lot of
speech recognition hardware idle at off peak periods. Typically the
off peak periods constitute the vast majority of the time.
Imperfect performance of the ASR systems means that there are costs
associated with incorrect recognition, and any improvement in
recognition accuracy is desirable.
BRIEF DESCRIPTION OF THE DRAWINGS
[0005] In order to describe the manner in which the above-recited
and other advantages and features of the disclosure can be
obtained, a more particular description of the principles briefly
described above will be rendered by reference to specific
embodiments thereof which are illustrated in the appended drawings.
Understanding that these drawings depict only exemplary embodiments
of the disclosure and are not therefore to be considered to be
limiting of its scope, the principles herein are described and
explained with additional specificity and detail through the use of
the accompanying drawings in which:
[0006] FIG. 1 illustrates an example system embodiment;
[0007] FIG. 2 is a functional block diagram that illustrates an
exemplary natural language spoken dialog system;
[0008] FIG. 3 illustrates an example method embodiment; and
[0009] FIG. 4 illustrates a set of speech recognition resources and
a plot of speech recognition demand.
DETAILED DESCRIPTION
[0010] The approach disclosed herein utilizes idle hardware to
improve the recognition performance with slower and larger
solutions that are impractical at peak load times, but are suitable
for off peak times, providing higher recognition accuracy. Speech
recognition applications typically include the necessary hardware
to handle projected or historical peak load conditions, given the
best ASR configuration in terms of accuracy vs. hardware load,
cost, latency, etc. trade offs. Those peak conditions are
relatively rare, and might occur once or twice a day or one or more
days a week, for example. Much of the time, much of the hardware is
idle. The ASR configuration used in speech recognition applications
can be configured differently for higher accuracy, but at the
expense of additional processing or additional cost. For example,
one approach combines output of two or more recognizers trained in
different ways to provide output that is more accurate then the
speech recognition output from any single recognizer alone. This
and other approaches in combination can utilize hardware more
effectively in off-peak times to provide more accurate recognition
output with virtually no additional cost in the application
provisioning. An ASR system can use a single more load intensive
solution or a collection of relatively less load intensive
solutions selectively for the most "needy" applications, or when
accuracy is most important as the CPU cycles become available. The
ASR system can make this decision based on what is more appropriate
and beneficial for the ASR task at hand.
[0011] In one aspect, an ASR system runs multiple recognizers in
parallel, each of which is trained with slightly different
characteristics, such as a different spectral analysis in the front
end, different pronouncing dictionaries, different training
algorithms, etc. In this manner, the ASR system produces an
improved recognition output by combining the individual recognition
outputs in any number of possible ways. This solution, although
more accurate than the best individual system, is also larger,
slower, and introduces marginally longer latencies. In general the
requirements of additional hardware to process the input speech
outweigh the benefit of this approach. However, if the speech
recognition hardware is already idle during off peak load periods,
the system can implement larger and more complex speech recognition
algorithms without incurring any additional cost besides a somewhat
more elaborate scheduling algorithm. The more elaborate scheduling
algorithm can be tailored to a particular combination of
applications running on the hardware platforms.
[0012] This approach can provide improved recognition accuracy
through more elaborate and computationally expensive ASR solutions
during off peak periods without the additional cost and/or
additional hardware when the available hardware is under-utilized
during off peak load times, which is most of the time.
[0013] Various embodiments of the disclosure are discussed in
detail below. While specific implementations are discussed, it
should be understood that this is done for illustration purposes
only. A person skilled in the relevant art will recognize that
other components and configurations may be used without parting
from the spirit and scope of the disclosure.
[0014] With reference to FIG. 1, an exemplary system 100 includes a
general-purpose computing device 100, including a processing unit
(CPU or processor) 120 and a system bus 110 that couples various
system components including the system memory 130 such as read only
memory (ROM) 140 and random access memory (RAM) 150 to the
processor 120. These and other modules can be configured to control
the processor 120 to perform various actions. Other system memory
130 may be available for use as well. It can be appreciated that
the disclosure may operate on a computing device 100 with more than
one processor 120 or on a group or cluster of computing devices
networked together to provide greater processing capability. The
processor 120 can include any general purpose processor and a
hardware module or software module, such as module 1 162, module 2
164, and module 3 166 stored in storage device 160, configured to
control the processor 120 as well as a special-purpose processor
where software instructions are incorporated into the actual
processor design. The processor 120 may essentially be a completely
self-contained computing system, containing multiple cores or
processors, a bus, memory controller, cache, etc. A multi-core
processor may be symmetric or asymmetric.
[0015] The system bus 110 may be any of several types of bus
structures including a memory bus or memory controller, a
peripheral bus, and a local bus using any of a variety of bus
architectures. A basic input/output (BIOS) stored in ROM 140 or the
like, may provide the basic routine that helps to transfer
information between elements within the computing device 100, such
as during start-up. The computing device 100 further includes
storage devices 160 such as a hard disk drive, a magnetic disk
drive, an optical disk drive, tape drive or the like. The storage
device 160 can include software modules 162, 164, 166 for
controlling the processor 120. Other hardware or software modules
are contemplated. The storage device 160 is connected to the system
bus 110 by a drive interface. The drives and the associated
computer readable storage media provide nonvolatile storage of
computer readable instructions, data structures, program modules
and other data for the computing device 100. In one aspect, a
hardware module that performs a particular function includes the
software component stored in a tangible and/or intangible
computer-readable medium in connection with the necessary hardware
components, such as the processor 120, bus 110, display 170, and so
forth, to carry out the function. The basic components are known to
those of skill in the art and appropriate variations are
contemplated depending on the type of device, such as whether the
device 100 is a small, handheld computing device, a desktop
computer, or a computer server.
[0016] Although the exemplary embodiment described herein employs
the hard disk 160, it should be appreciated by those skilled in the
art that other types of computer readable media which can store
data that are accessible by a computer, such as magnetic cassettes,
flash memory cards, digital versatile disks, cartridges, random
access memories (RAMs) 150, read only memory (ROM) 140, a cable or
wireless signal containing a bit stream and the like, may also be
used in the exemplary operating environment. Tangible
computer-readable storage media expressly exclude media such as
energy, carrier signals, electromagnetic waves, and signals per
se.
[0017] To enable user interaction with the computing device 100, an
input device 190 represents any number of input mechanisms, such as
a microphone for speech, a touch-sensitive screen for gesture or
graphical input, keyboard, mouse, motion input, speech and so
forth. The input device 190 may be used by the presenter to
indicate the beginning of a speech search query. An output device
170 can also be one or more of a number of output mechanisms known
to those of skill in the art. In some instances, multimodal systems
enable a user to provide multiple types of input to communicate
with the computing device 100. The communications interface 180
generally governs and manages the user input and system output.
There is no restriction on operating on any particular hardware
arrangement and therefore the basic features here may easily be
substituted for improved hardware or firmware arrangements as they
are developed.
[0018] For clarity of explanation, the illustrative system
embodiment is presented as including individual functional blocks
including functional blocks labeled as a "processor" or processor
120. The functions these blocks represent may be provided through
the use of either shared or dedicated hardware, including, but not
limited to, hardware capable of executing software and hardware,
such as a processor 120, that is purpose-built to operate as an
equivalent to software executing on a general purpose processor.
For example the functions of one or more processors presented in
FIG. 1 may be provided by a single shared processor or multiple
processors. (Use of the term "processor" should not be construed to
refer exclusively to hardware capable of executing software.)
Illustrative embodiments may include microprocessor and/or digital
signal processor (DSP) hardware, read-only memory (ROM) 140 for
storing software performing the operations discussed below, and
random access memory (RAM) 150 for storing results. Very large
scale integration (VLSI) hardware embodiments, as well as custom
VLSI circuitry in combination with a general purpose DSP circuit,
may also be provided.
[0019] The logical operations of the various embodiments are
implemented as: (1) a sequence of computer implemented steps,
operations, or procedures running on a programmable circuit within
a general use computer, (2) a sequence of computer implemented
steps, operations, or procedures running on a specific-use
programmable circuit; and/or (3) interconnected machine modules or
program engines within the programmable circuits. The system 100
shown in FIG. 1 can practice all or part of the recited methods,
can be a part of the recited systems, and/or can operate according
to instructions in the recited tangible computer-readable storage
media. Generally speaking, such logical operations can be
implemented as modules configured to control the processor 120 to
perform particular functions according to the programming of the
module. For example, FIG. 1 illustrates three modules Mod1 162,
Mod2 164 and Mod3 166 which are modules configured to control the
processor 120. These modules may be stored on the storage device
160 and loaded into RAM 150 or memory 130 at runtime or may be
stored as would be known in the art in other computer-readable
memory locations.
[0020] FIG. 2 is a functional block diagram that illustrates an
exemplary natural language spoken dialog system. Spoken dialog
systems aim to identify intents of humans, expressed in natural
language, and take actions accordingly, to satisfy their requests.
Natural language spoken dialog system 200 can include an automatic
speech recognition (ASR) module 202, a spoken language
understanding (SLU) module 204, a dialog management (DM) module
206, a spoken language generation (SLG) module 208, and
synthesizing module 210. The synthesizing module can be any type of
speech output module. For example, it can be a module wherein one
prerecorded speech segment is selected and played to a user. Thus,
the synthesizing module represents any type of speech output. The
present disclosure focuses on innovations related to the ASR module
202 and can also relate to other components of the dialog
system.
[0021] The ASR module 202 analyzes speech input and provides a
textual transcription of the speech input as output. SLU module 204
can receive the transcribed input and can use a natural language
understanding model to analyze the group of words that are included
in the transcribed input to derive a meaning from the input. The
role of the DM module 206 is to interact in a natural way and help
the user to achieve the task that the system is designed to
support. The DM module 206 receives the meaning of the speech input
from the SLU module 204 and determines an action, such as, for
example, providing a response, based on the input. The SLG module
208 generates a transcription of one or more words in response to
the action provided by the DM 206. The synthesizing module 210
receives the transcription as input and provides generated audible
speech as output based on the transcribed speech.
[0022] Thus, the modules of system 200 recognize speech input, such
as speech utterances, transcribe the speech input, identify (or
understand) the meaning of the transcribed speech, determine an
appropriate response to the speech input, generate text of the
appropriate response and from that text, generate audible "speech"
from system 200, which the user then hears. In this manner, the
user can carry on a natural language dialog with system 200. Those
of ordinary skill in the art will understand the programming
languages for generating and training ASR module 202 or any of the
other modules in the spoken dialog system. Further, the modules of
system 200 can operate independent of a full dialog system. For
example, a computing device such as a smartphone (or any processing
device having a phone capability) can include an ASR module wherein
a user says "call mom" and the smartphone acts on the instruction
without a "spoken dialog." A module for automatically transcribing
user speech can join the system at any point or at multiple points
in the cycle or can be integrated with any of the modules shown in
FIG. 2.
[0023] Having disclosed some basic system components, the
disclosure now turns to the exemplary method embodiment shown in
FIG. 3. For the sake of clarity, the method is discussed in terms
of an exemplary system 100 such as is shown in FIG. 1 configured to
practice the method.
[0024] FIG. 3 illustrates a computer-implemented method for
improving automatic speech recognition performance. A system 100
identifies idle speech recognition resources (302). Idle speech
recognition resources can include networked computing devices,
spare CPU cycles, available memory or storage, bandwidth, available
throughput on a local bus, etc. In one aspect, idle resources are
not truly idle, but operate in a superfluous way. For example, if
two speech recognition resources are recognizing speech together
for a particular speaker, and one of the two is able to acceptably
recognize the speech by itself, the other can appropriately be
termed "idle" even though it is actively engaged in speech
recognition. In this situation, the system 100 can reallocate
underutilized, superfluous, or otherwise unnecessary resources from
their current recognition task to another recognition task. The
system 100 can allocate various resources in a single computing
device to different recognition tasks.
[0025] The system 100 establishes one or more supplemental speech
recognizer on the idle resources based on overall speech
recognition demand (304). The supplemental speech recognizer can
differ from a main speech recognizer. The supplemental speech
recognizer and the main speech recognizer can be associated with a
particular speaker, group, or class of speakers. A scheduling
algorithm can tailor a particular combination of speech recognition
resources. The scheduling algorithm can also release the
supplemental speech recognizer based on increased demand for speech
recognition. Along the lines of the supplemental speech recognizer,
the system can establish extra speech recognizers tailored to
speech which requires additional accuracy as extra speech
recognition resources become idle. For example, the system 100 is
assigned to recognize speech that is very difficult, but
insufficient resources are currently idle. As new resources become
idle or available, the system 100 can establish extra speech
recognizers before or during the recognition process to assist in
the difficult recognition task. In the case of multiple
supplemental speech recognizers, each can differ from each other
based on one or more of spectral analysis in a front end,
pronouncing dictionaries, and/or training algorithms.
[0026] The system 100 performs speech recognition on speech
received from the particular speaker in parallel with the main
speech recognizer and the supplemental speech recognizer (306) and
combines results from the main and supplemental speech recognizer
(308). The system 100 then recognizes the received speech based on
the combined results (310). The system 100 can return the
supplemental speech recognizer to an idle state after recognizing
all or some of the received speech. In some cases where the system
100 is uncertain whether to expect additional speech, the system
100 can incrementally release supplemental speech recognizers to an
idle state over a determined period of time.
[0027] FIG. 4 is an exemplary chart 400 of available speech
recognition resources 402 and a plot of projected speech
recognition demand 404. The chart 400 is divided into weekdays. In
this system, the available speech recognition resources are
constant. Typically a speech recognition system designer accounts
for the projected peak loads and plans the speech recognition
system around those projections. As shown here, a typical demand
projection 404 only rarely approaches peak demand, in this case
only on Fridays. A substantial portion of the speech recognition
resources go unused and remain in an idle state in a typical speech
recognition system. During periods of low demand such as the low
points between days or pretty much any time other than peak load
times, a speech recognition system 100 practicing the method
disclosed herein can dynamically establish a main speech recognizer
and one or more supplemental speech recognizers on the idle
resources dedicated to a particular speaker. The system can then
process speech from the speaker in parallel with the main speech
recognizer and the supplemental speech recognizers, combine results
from the main and supplemental speech recognizers, and recognize
the received speech based on the combined results. In this way, the
speech recognition system 100 can dynamically scale the complexity
of the recognition based on overall speech recognition demand. A
dynamic spoken dialog system adapts based on needs of particular
callers and available speech recognition resources, such as
processor cycles, memory, storage space, bandwidth, and so
forth.
[0028] In one aspect, speech recognizers run faster than real time
to provide a level of natural language responsiveness that users
expect. The system 100 can dynamically scale some speech
recognition properties to run faster or slower based on demand. For
example, the system 100 can increase the beam from its nominal
value "B" to 1.1*B or 1.2*B to increase recognition accuracy at
higher computational cost, while still running at faster speeds
than real-time, or reduce the load by using smaller beam of 0.9*B
or 0.8*B at the expense of lowered recognition accuracy. The system
100 can introduce larger language models into an existing speech
recognizer, which requires increased amounts of memory. The system
can allocate additional parallel speech recognizers and combine
their recognition results. The system 100 can use different
recognizers to provide recognition strings from word lattices and
confusion networks which can then be combined using weighted voting
(commonly referred to as ROVER-ing). For example, one recognizer
can optimize at the sentence level and another recognizer can
optimize at the word level. The system 100 can recognize speech
using dialect specific models in parallel with generic speech
models. The system 100 can use one or more of these approaches,
depending on a user profile, speech characteristics, overall
demand, cost, latency, projected demand, and/or other factors. In
other words, the system 100 can adapt along a range of complexity
that scales based on demand within with available resources and
also takes in to account specific user needs.
[0029] In one aspect, rather than scaling complexity up, the system
100 scales complexity down to meet demand which exceeds the
projected peak and the available speech recognition resources. For
example, during such exceptionally high load times, the system can
sacrifice 10% accuracy if the decreased accuracy only requires 1/3
as many resources. The system 100 can selectively decrease accuracy
for speakers whose speech the system already recognizes very well
with few errors, for example. The system can even shave some
resources away from the best performing speakers and reallocate
them to provide additional help for borderline or needy speakers.
In this way, the decrease in accuracy is not as perceptible for
these speakers as it would be for speakers whose speech the system
recognizes with difficulty. The system 100 can dynamically scale
complexity down and back to normal levels as overall demand returns
below the available resources.
[0030] Embodiments within the scope of the present disclosure may
also include tangible computer-readable storage media for carrying
or having computer-executable instructions or data structures
stored thereon. Such computer-readable storage media can be any
available media that can be accessed by a general purpose or
special purpose computer, including the functional design of any
special purpose processor as discussed above. By way of example,
and not limitation, such computer-readable media can include RAM,
ROM, EEPROM, CD-ROM or other optical disk storage, magnetic disk
storage or other magnetic storage devices, or any other medium
which can be used to carry or store desired program code means in
the form of computer-executable instructions, data structures, or
processor chip design. When information is transferred or provided
over a network or another communications connection (either
hardwired, wireless, or combination thereof) to a computer, the
computer properly views the connection as a computer-readable
medium. Thus, any such connection is properly termed a
computer-readable medium. Combinations of the above should also be
included within the scope of the computer-readable media.
[0031] Computer-executable instructions include, for example,
instructions and data which cause a general purpose computer,
special purpose computer, or special purpose processing device to
perform a certain function or group of functions.
Computer-executable instructions also include program modules that
are executed by computers in stand-alone or network environments.
Generally, program modules include routines, programs, components,
data structures, objects, and the functions inherent in the design
of special-purpose processors, etc. that perform particular tasks
or implement particular abstract data types. Computer-executable
instructions, associated data structures, and program modules
represent examples of the program code means for executing steps of
the methods disclosed herein. The particular sequence of such
executable instructions or associated data structures represents
examples of corresponding acts for implementing the functions
described in such steps.
[0032] Those of skill in the art will appreciate that other
embodiments of the disclosure may be practiced in network computing
environments with many types of computer system configurations,
including personal computers, hand-held devices, multi-processor
systems, microprocessor-based or programmable consumer electronics,
network PCs, minicomputers, mainframe computers, and the like.
Embodiments may also be practiced in distributed computing
environments where tasks are performed by local and remote
processing devices that are linked (either by hardwired links,
wireless links, or by a combination thereof) through a
communications network. In a distributed computing environment,
program modules may be located in both local and remote memory
storage devices.
[0033] The various embodiments described above are provided by way
of illustration only and should not be construed to limit the scope
of the disclosure. Those skilled in the art will readily recognize
various modifications and changes that may be made to the
principles described herein without following the example
embodiments and applications illustrated and described herein, and
without departing from the spirit and scope of the disclosure.
* * * * *